The disclosure is directed to an object detection method using a CNN model and an object detection apparatus thereof. In an aspect, the object detection method includes generating a sensor data; processing the sensor data by using a first object detection algorithm to generate a first object detection result; processing the first object detection result by using a plurality of stages of sparse update mapping algorithm to generate a plurality of stages of updated first object detection result; processing a first stage of the stages of updated first object detection result by using a plurality of stages of spatial pooling algorithm between each of stages of sparse update mapping algorithm; executing a plurality of stages of deep convolution layer algorithm to extract a plurality of feature results; and performing a detection prediction based on a last-stage feature result.
Legal claims defining the scope of protection, as filed with the USPTO.
1. An object detection method using a CNN model, the method comprising: generating, by using a sensor, a sensor data; processing the sensor data by using a first object detection algorithm to generate a first object detection result, wherein the first object detection result includes bounding box information; processing the first object detection result by using a plurality of stages of sparse update mapping algorithm according to the bounding box information to generate a plurality of stages of updated first object detection result in the CNN model; processing a first stage of the stages of updated first object detection result by using a plurality of stages of spatial pooling algorithm to reduce the spatial dimensions of the stages of updated first object detection result between each of stages of sparse update mapping algorithm; executing a plurality of stages of deep convolution layer algorithm to correspondingly extract a plurality of feature results based on the stages of updated first object detection result in the CNN model; and performing a detection prediction based on a last-stage feature result generated by a last stage of the stages of deep convolution layer algorithm.
2. The object detection method of claim 1 , wherein each of the stages of sparse update mapping algorithm comprising: generating a box information map with a plurality of bounding boxes as the updated first object detection result by mapping the bounding boxes information to the box information map, wherein the box information map comprised by a plurality of grid cells, and numbers of grid cells in the box information map are the same as numbers of pixels in the sensor data.
3. The object detection method of claim 2 , wherein each of the stages of sparse update mapping algorithm further comprising: setting the grid cells overlapped with bounding boxes as a first number region and setting the grid cells without the bounding boxes as a second number region; and setting a value of the first number region as the first index and setting a value of the second number region as the second index, wherein the first index and the second index are different.
4. The object detection method of claim 3 , wherein the first index is 1 and the second index is 0.
5. The object detection method of claim 3 , wherein each of the stages of sparse update mapping algorithm further comprising: combining any set of the bounding boxes that are spatially overlapping as a singular collection of overlapping bounding boxes.
6. The object detection method of claim 1 , wherein each of the stages of deep convolution layer algorithm comprising: processing a pre-stage updated first object detection result of the stages of updated first object detection result based on a pre-stage feature result of the feature results by using a position layer algorithm to generate a strip feature map; processing the strip feature map by using a convolution layer algorithm to generate a convoluted strip feature map; processing the convoluted strip feature map based on a prior-time feature result from an earlier tome point of the stages of deep convolution layer algorithm by using a recovery layer algorithm to generate the feature result; and processing the feature result by using a pooling layer algorithm to reduce the spatial dimension of the feature result.
7. The object detection method of claim 6 , wherein the position layer algorithm comprising: receiving the pre-stage updated first object detection result and the pre-stage feature result; dividing the pre-stage updated first object detection result based on the pre-stage feature result to generate a position map with a plurality of image regions, wherein each of the image regions assembled by N*N grid cells, and N is a positive integer; setting a value of the image regions overlapped with the first number region as the first index; and convoluting the position map to the strip feature map.
8. The object detection method of claim 7 , wherein the convolution layer algorithm further comprising: convoluting the strip feature map to generate a convoluted strip feature map.
9. The object detection method of claim 8 , wherein the recovery layer algorithm further comprising: receiving the convoluted strip feature map; and operating the convoluted strip feature map and the prior-time feature result to generate the feature result.
10. The object detection method of claim 1 , wherein the step of processing the sensor data by using a first object detection algorithm comprising: detecting the sensor data to find a region of interest based on a computer vision; and generating the bounding box information according to the region of interest.
11. An object detection apparatus comprising: a sensor for generating a sensor data; and a processor coupled to the sensor and configured at least for: processing the sensor data by using a first object detection algorithm to generate a first object detection result, wherein the first object detection result includes bounding box information; processing the first object detection result by using a plurality of stages of sparse update mapping algorithm according to the bounding box information to generate a plurality of stages of updated first object detection result in the CNN model; processing a first stage of the stages of updated first object detection result by using a plurality of stages of spatial pooling algorithm to reduce the spatial dimensions of the stages of updated first object detection result between each of stages of sparse update mapping algorithm; executing a plurality of stages of deep convolution layer algorithm to correspondingly extract a plurality of feature results based on the stages of updated first object detection result in the CNN model; and performing a detection prediction based on a last-stage feature result generated by a last stage of the stages of deep convolution layer algorithm.
12. The object detection apparatus of claim 11 , wherein each of the stages of sparse update mapping algorithm comprising: generating a box information map with a plurality of bounding boxes as the updated first object detection result by mapping the bounding boxes information to the box information map, wherein the box information map comprised by a plurality of grid cells, and numbers of grid cells in the box information map are the same as numbers of pixels in the sensor data.
13. The object detection apparatus of claim 12 , wherein each of the stages of sparse update mapping algorithm further comprising: setting the grid cells overlapped with the bounding boxes as a first number region and setting the grid cells without the bounding boxes as a second number region; and setting a value of the first number region as the first index and setting a value of the second number region as the second index, wherein the first index and the second index are different.
14. The object detection apparatus of claim 13 , wherein the first index is 1 and the second index is 0.
15. The object detection apparatus of claim 13 wherein each of the stages of sparse update mapping algorithm further comprising: combining any set of the bounding boxes that are spatially overlapping as a singular collection of overlapping bounding boxes.
16. The object detection apparatus of claim 11 , wherein each of the stages of deep convolution layer algorithm comprising: processing a pre-stage updated first object detection result of the stages of updated first object detection result based on a pre-stage feature result of the feature results by using a position layer algorithm to generate a strip feature map; processing the strip feature map by using a convolution layer algorithm to generate a convoluted strip feature map; processing the convoluted strip feature map based on a prior-time feature result from an earlier tome point of the stages of deep convolution layer algorithm by using a recovery layer algorithm to generate the feature result; and processing the feature result by using a pooling layer algorithm to reduce the spatial dimensions of the feature result.
17. The object detection apparatus of claim 16 , wherein the position layer algorithm comprising: receiving the pre-stage updated first object detection result and the pre-stage feature result; dividing the pre-stage updated first object detection result based on the pre-stage feature result to generate a position map with a plurality of image regions, wherein each of the image regions assembled by N*N grid cells, and N is a positive integer; setting a value of the image regions overlapped with the first number region as the first index; and convoluting the position map to the strip feature map.
18. The object detection apparatus of claim 17 , wherein the convolution layer algorithm further comprising: convoluting the strip feature map to generate a convoluted strip feature map.
19. The object detection apparatus of claim 18 , wherein the recovery layer algorithm further comprising: receiving the convoluted strip feature map; and operating the convoluted strip feature map and the prior-time feature result to generate the feature result.
20. The object detection apparatus of claim 11 , wherein the processor is configured at least for processing the sensor data by using a first object detection algorithm comprising: detecting the sensor data to find a region of interest based on a computer vision; and generating the bounding box information according to the region of interest.
21. The object detection apparatus of claim 11 , wherein the sensor comprises a camera.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
December 11, 2018
August 18, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.